Cost review

Newest SDK:

Espressif has moved the SDK to GitHub.  At this point, It is version 2.1.  I have no indication that they have done anything with the ADC Feature I requested.  I have not updated to this newest SDK yet. I will do that before I spend any time working on firmware.

New component costing:

The 74LVC2T45 level shifter chips are US $0.32 each in low quantities.  The PCA9306 I2C level shifter are about US $0.60 each in low quantities.  This results in the costs of the level shift components arount $1.50.  The original single part costs around US $1.25 each in low quantities.  This is not a big difference in cost but it does affect board space.

Design review:

Continue reading

Schematic update

Arduino compass toy:

I have had a little time to play with the OLED panel this week. Unfortunately, I still can’t get it to work.  I have been trying to get it to work in SPI mode, But I think I now want to try to start working in either 8080 or 6800 modes.

Schematic Update:

I updated the schematic to reflect the results of my testing from 2 weeks ago.

I still need to try a high gain transistor in place of the one I used to see what will happen.

I am waiting to make some decisions before I update the schematic on GitHub.


Continue reading

Level Shifter Research


My ADC feature request on Espressif’s BBS has had 118 views. Unfortunately no word from the developers.  This might mean I have to add an ADC chip to the design.

The problem:

I’ve been thinking about the level shifter and why it was causing my software to crash. 10M Ohm and 10pF capacitance are very small loads. The software crashing just because a scope probe is connecting is very bad.(stating the obvious)

The hypothesis:

Bidirectional level shifters are typically designed to react quickly to transitions and drive very weakly in the steady state. With the target pins unconnected the shifter is steady state and the scope probe might be enough to cause a problem.

The Experiment:

Continue reading

Layout check … continued

With the new check plots I made last week, I checked the layout net by net like I did back in Sept 5, 2015. I am still uncertain about the pinout of the ESP-12F modules. To test it, I removed the ESP-12E installed on the board I am currently using, and replaced it with a new ESP-12F.  I had to be careful that the jumper wires all got back to where they were before removing the module.

The new module programmed first try but didn’t boot correctly.  I had forgot to upload default values, the flash comes programmed for a different memory layout.  Once I installed the default settings, it booted up correctly.  I had removed the test for HSPI/SPI RAM, so I went back into the code and re-enabled the Hello World ram test.

I went into serialInit() and changed the line DISPLAY_MENU(); to DISPLAY_MENU_W_SPI(); Re-compiled and tested.  Unfortunately the result was not “Hello World”. When I cleaned up the code, I made the write to SPI RAM easier, but I need to initialize the hspi.  I started connecting my mixed signal oscope to analyze what was happening on the SPI bus.  And… I found a loose jumper wire.  The one that connects the chip select between the ESP-12F and the SPI RAM. Once this was connected, the extra stuff I had put into the software became unnecessary.  I deleted the extra code and it works just fine.  The firmware is back to where it was before I started testing. There are no changes to upload.

This means that the circuit is tested and the current layout is ready to send to fabrication.  I zipped up the gerber files. Then checked pricing with PCBway, DFRobot, and Seeed Studio.  With shipping PCBway came in a little less at US $19.  With PCBway, I have to wait for a design review before they will give me a final price and allow me to pay for it. Usually this is pretty quick but depends on the day it is submitted.

Have you bought boards from Chinese manufacturers? How quickly did you get them?

Sigma Delta Testing V00A

I have been fighting using the Sigma Delta(ΣΔ) for a few weeks now and all of my research so far has not been fruitful.  So, My intention this week is to implement a very basic ΣΔ operation that will change the prescaler for the whole range of possibilities 0 – 255.

The hardware watchdog timer triggers after about 6 seconds without a WDT reset. I have decided I want the sigma delta prescaler to step from 0 to 255 every 5 seconds.  I will put each step of the prescaler out to the serial port using system calls.

I will put all the code in the user_main.c file so that I, and anyone else that looks at my code can see all that is going on.  I extracted a new set of SDK files into a new folder called SigmaDelta.  I then created a new Eclipse project file in this new folder.    I struggled with creating the project in Eclipse.  I finally got it set up the way I wanted.  In the past setting up a project in Eclipse has always been easier, but this time it just didn’t work well for me.

I copied my shell scripts(that is batch files for windows users) that I use to automate building and testing from the UProgrammer folder. Next, I removed the references to AT_CUSTOM_UPGRADE in user_main.c and user_config.h.  I then performed a sanity check by putting one line of code to send “Hello World” out the serial connection and tried to build.  It failed because I forgot a semicolon.  I fixed that and got a new message saying code will not fit in irom0_0_seg.  I need to update the linker  file.  I copied the one from the UProgrammer folder.  Built again and it compiled. Next I uploaded it to see if it still works.  It doesn’t seem to work.  The baud rate is set to 115200, but I am getting a little bit of garbled text after boot completes.  I decided to try putting a small delay(5 ms) in before printing the “hello World”. This also didn’t work.

I need to implement a timer and looked at the sntp example in the API guide.  I set the timer to repeat every 5 seconds. Unfortunately it seems to just hang. I had a similar problem when I moved to this version of the API, I removed the uart_init line and I started getting regular data coming across the serial line. Still garbled, but happening every 5 seconds.  So I looked at uart_init from the uart.c. I found and copied the call that actually sets the baud rate.  I used this to modify the baud rate for UART0 and I started getting Hello World every 5 seconds.  Next step, turn on the Sigma Delta and start incrementing the prescaler.

I created the variables prescaler and duty and set them to 0. Then I went back to my old code and grabbed a copy of the set prescaler code.  I put the code in to set the prescaler, increment it, and print out the value every the timer event is called. I compiled, uploaded it and I saw numbers increasing across the serial terminal.  Then I waited for it to get to 100.  It continued to count.  I let it continue all the way up to 255 and then changed the wifi mode to STATION_MODE like the UProgrammer firmware. Next, I modified the code to turn on the output(GPIO4) and change the duty cycle.

Still working. Connected to wifi and tested again. Still working.  I explicitly set GPIO4 as an output set to 1 and tried again. Still working I turned on HSPI overlap mode and tried again. It started crashing and rebooting.  It looks like somehow the HSPI is conflicting with the SigmaDelta generator.

I commented out the code to set up GPIO4, and the code to connect to the wifi access point and it crashes. I changed the duty back to 0 and it doesn’t crash anymore. Next I checked the Pin assignment worksheet to see what else is connected to GPIO4 and I didn’t find anything interesting.

It looks like I have to do an HSPI overlap deinit immediately after getting data out of the RAM.  This could be a problem because I want to be generating Vpp while transferring data into the processor.  This could limit my options.  If my code is small, I can generate Vpp. If it’s large, I may have to program in blocks.

I do have the capabilities to assemble and test the voltage booster now.  I will go back into the UProgrammer Firmware code and deinit the HSPI overlap and see what happens.

I have put this week’s test code on GitHub here. How do you find problem code?

Update to SDK 1.5.4 Firmware V00J

Since I don’t seem to be making any progress on the Sigma Delta output, I decided to update my copy of Espressif’s API. I went to and clicked on the SDKs thread under the heading DOWNLOADS. In the list i looked for the most recent Non OS SDK release.  I saw that ESP8266_NONOS_SDK_V1.5.4_16_05_20 is the most recent release. It’s file name indicates a release date of May 20th 2016.  I downloaded the zip file and extracted the contents.

To Make moving forward a little easier, I renamed the SDK folder to UProgrammer.  This will help me maintain paths the next time I move to a new SDK Version.  Each time I update to a new SDK, I will Rename the OLD SDK folder to indicate it’s version number, and rename the new SDK folder to UProgrammer.  To maintain compatibility with Github’s folder naming, I will continue to keep the UProgrammer-Firmware folder name for the project files.  This folder will be sitting in the SDK folder I just renamed to UProgrammer.  I then created an empty text file with the name of the SDK file to mark the version I am working in.  Then I copied the hidden file .metadata from the old SDK folder into the new (UProgrammer) SDK folder.

I opened eclipse and pointed the workspace to the UProgrammer folder and it loaded the files as if I hadn’t changed anything. Then I made edited the comment block at the top of the file to indicate I had moved to NONOS SDK V1.5.4.  I saved the file and then checked the file in the new directory structure for the changes. It has the change I had just made.  To make sure there wasn’t something weird going on, I also checked the file in the old directory structure. It was unchanged.  I had planned on dealing with more problems just in case they happened.

I compiled the code to see if anything broke moving to the new SDK.  Things definitely broke. First the files all compiled but linking failed.  The first error message was:

section `.irom0.text' will not fit in region `irom0_0_seg'

This reminded me that I modified the linker file to allow for more room for the code space in a 4MByte flash chip.  I made the modification to the new file and saved it.

Modified linker file

I tried compiling again.  Again it failed to link but I was no longer getting the memory failure. This time the first error message was:

 undefined reference to `uart0_sendStr'

I didn’t have the most recent programming reference, so I went back to to download it.  I got the 2C-ESP8266_SDK_API Guide_EN_1.5.4.pdf.  It turns out uart0_sendStr() isn’t in the documentation.  So I looked in my code to find it.  I found the prototype for it in uart.h. but it doesn’t appear anywhere else. This makes me wonder how I was compiling last week.  I did find it in uart.c in the drivers folder.  At one time that file was part of the project. I am guessing the object file was still getting linked in to the project.  The compiled object files are in hidden folders and didn’t get copied by default.  (It’s a good thing.  The versions of code on Github would not have worked. Now they will get fixed.)  I copied The code for uart0_sendStr() into simple_serial.c and recompiled expecting it to fail because uart0_sentStr() calls something else in uart.c.

Again I looked at the first error from the linker which was:

../lib/libat.a(at_ipCmd.o): In function `at_set_rx_buf_state':
(.irom0.text+0x10ac): undefined reference to `espconn_secure_sent'

I am not using any secure features or any AT functions, so I decided to look at the make files. I used a Linux tool called meld to compare files.  There were very small differences.  Looking at my makefile, I noticed four lines that effectively added the -lat linker option.  I commented out those 4 lines and recompiled.  Now the linker only has 4 errors remaining, all of which can be fixed by copying code from the uart.c file.  After copying each function I needed from uart.c, I finally got a successful compile of the code.  I then uploaded it to the PCB to see if it still runs.  It ran the wrong code.  I now get a message in the serial terminal “dmode : softAP (XXX) indicating it is booting up as an access point.

I was able to connect to the device and dhcp worked, but an attemp to connect by http was refused.  Since the messages are system messages I decided to try a UART Swap. No change in the behavior.  I need to do a sanity check, this is a new API so back to doing a Hello World test. I finally got confirmation that my code is what is being uploaded to the module.  I placed the following function call:

system_set_os_print(0); // Turn off System Messages

into the code. I am no longer getting any output in the terminal.  I wasn’t getting any error from compiling but it turns out the function uart_tx_one_char() needed to be copied from the uart.c to simple_serial.c I chose not to have uart.c to reduce code size, there is a lot of code in uart.c that I will never use.

I cleaned up some of the things I tested recompiled and posted a new copy to Github. Do you have any tips for migrating between environments? I’d love to hear about them.

Sigma Delta still causing resets.

When I first built this revision of PCB, the board would reboot a lot.  That turned out to be the mis-wiring of the SPI RAM.  Part of that process in trouble shooting that board, I had guessed that I had a noisy power connection to the Wi-Fi module. So one thing that I had tried was adding a large reservoir capacitor and a small filter capacitor near the power pins of the module. Unless I am drawing too much current from the voltage regulator, this should eliminate any possible power issues.

This means my first test is to measure how much current is being drawn from the voltage regulator.  To do that I removed the jumper that bypassed the lithium cell charger and attached 3.7V to the PCB from a bench power supply.  To do this I removed all the test pins I had soldered to the module and soldered one to +BATT and another one to GND.  I connected my meter in series with the power supply and turned it all back on.  I measured the current consumed between 75 and 80 mA for the whole range of 0 to 99 prescaler.  This measurement was made with the radio inactive.

I think I can rule out the power supply as my problem.  Next I decided to change the software to take steps of 10 instead of 1 for adjusting the prescaler.  If the failure is because of an accumulated error, this should happen later and I should be able to get a number larger than 100 without a failure. I re-soldered the  bypass of the Lithium Cell charger and uploaded the change. I got to 91 No Failure. At 101 it still failed.  I change the step to 30 and when I got to 120, I got a reset cause:2. (maybe there is a clue here, This means external reset.)  I found this in the 2c-ESP8266_SDK_Programming Guide.pdf file from the Espressif BBS.

There are two pins on the module that can be used for external reset.  Since the serial to usb bridge is controlling the reset line, I decided to look into the EN pin first. It looked good with the right resistor in place to pull it up.  I then started looking through the code to make sure the Sigma Delta was configured to the correct pin.  Turns out that it wasn’t configured to output on any pin.  There was a missing statement in the code setting pin4 to output the sigma delta channel.  I added that line back in, saved, recompiled, uploaded and tested again. It still is failing.

Going back to the external reset, I decided to put lower value pullup resistors on the reset lines. I changed them from 10 K to 1 K.  This should reduce any affect of noise on the pins and still not draw too much current when pulled low. It still fails in the same way.  I have another project where the sigma delta is working very well.  The main difference I can see is that on this project I am reading the sigma delta twice before I write to it. In my other project, it reads the register once.  So I am going to make a simple uint8 variable to retain the value I am writing to the Sigma Delta prescaler.

It is still not fixed.  Have you seen this problem? Have you solved it?



Switching to SPI from HSPI Firmware V00I

I decided to try to use the SPI module instead of overlap mode this week.  I went into the code that writes and reads from the SPI and changed the working registers to SPI in place of HSPI. I had to store and restore 6 variables in the writeRam() and readRam() functions.  It worked but sent more data than expected. I noticed that I left the overlap init code in place, I removed it and tried again. I still got more data than I expected. I searched for HSPI and found one place that I had forgot to change it to SPI. I fixed that and now it is working. The code is also working without crashing.

SPI Write Hello World

Next I added the read function back into the main code to see what would happen. I saw only 4 bytes on the SPI bus. I had missed a value change I needed to make the SPI read 12 bytes.    It seems that either the DMA or an Interrupt routine is taking control of the SPI Hardware before i can finish a read. I disabled interrupts and it’s still only sending the command (3) and the address (0) plus 6 more clock cycles.

Incomplete SPI Read

Since SPI is used for loading code into internal RAM as well as running it directly from flash, it might be difficult to figure out when I can use it for the SPI  RAM.  I had the HSPI appearing to work in overlap mode.  Since I can switch between overlap and normal mode on the fly, maybe the HSPI in overlap mode is the best choice. In the writeRam() and readRam() functions, I changed the all references to SPI Registers to HSPI Registers and I commented out the code to save and restore the SPI Registers.  I used find and replace to simplify the changes.  I searched on (SPI) and filled in the replace with (HSPI) and then I selectively hit REPLACE/FIND for each instance.

Then I built and tested the code, hopefully back to where I was last week.  I got all zeros but I hadn’t done a write since I started working on it this time. So I put the writeRam() just before the readRam() in user_init() and tried again. On the logic analyzer, the write looks exactly like I would expect. The read is still not getting anything but zeros for the data section.  Next step is to look at the HOLD and WP lines on the SPI RAM. I had to look at the datasheet to see what state they needed to be in to work correctly.  Looking at the datasheet, there is no WP line, but there is a HOLD line.  If HOLD is low, the serial interface of the chip doesn’t work.  I clipped one more logic analyzer line on to Pin 7 of the SPI RAM.  The signal looks identical to MOSI.  I should have cut the trace when I was re-wiring this board but I missed it. I proceeded to cut it and add a jumper from the HOLD line to the ESP12E Pin 11. this is my best guess for the pins to use SQI. With Re-testing I got a successful read from the chip.

SPI Successful read

Next week, I need to verify the SPI working registers have the expected values from the ram.  I put the current version of the code up on GitHub.

How do you handle cutting in circuit changes?  Do you have any stories about a design change only to find it best to go back to the original plan?

Understanding HSPI Firmware V00H

What I wasn’t understanding last week is the definition of HSPI_CS.  It is the same as CS0 in the HSPI module.  So when I turn on overlap mode, of course, it maps to SPI_CS0.  That leaves me with setting the HSPI to use CS1(GPIO1/U0TXD) or CS2(GPIO0).  Since I use U0TXD for reprogramming the module, CS1 is unavailable.

I cut the trace going to U2 CS0 and added a jumper wire to GPIO0.  I then changed the code to use CS2 to enable the RAM.  This included setting the HSPI to use CS2 and setting GPIO0 MUX to be SPICS2.  And I got what looks like a successful write to the SPI RAM.

Successful SPI Write

The byte command (2) followed by 24bit address (0) followed by the data.  The first four bytes of data are “lleH”.  This is backwards to what I was planning, I can change it by setting a bit or changing the bytes I put into the registers. As long as I read them back the same way, it really doesn’t matter as long as I read and write aligned to 4 byte words.  The data is one byte short, I had a typo sending only 11 bytes of data.

Note: When data is aligned to 4 bytes, it means you need to read or write only on increments of 4 bytes.  So you would only Write or Read to/from addresses that in hexadecimal end in 0, 4, 8, or C. Since this chip is a 32 bit processor, this is required often.

I then decided to implement the hard coded read command.  I copied the write code, changed the command from 2 to to 3, set the HSPI to read back  and zeroed out the working registers so that I can see if they have been filled with the data I stored in the RAM. Then I commented out the writeRam() call in user_init and added readRam() in it’s place. Doing this without powering down, I should get the data back without having to write it again. I forgot to switch the length to the receive length register; I fixed this then I got back all zeros for the data.

Unsuccessful SPI read

I see the command (3) followed by the address (0) followed by 12 Bytes of data(0). So I have more debugging to do.  My first guess is the WP or HOLD lines of the chip are not doing what I need them to do.

I have gotten a better understanding of the SPI hardware on the ESP8266.  It looks to me that I can get the same control by just using the main SPI channel to control my RAM chip.  This needs testing.  I will need to save any registers I change and restore them when I am done accessing the RAM.  This will also have to be done from code that runs from internal RAM.

I have put the current state of the code on GitHub.  Use the link under the search box on the right side of this page.

Have you used chips with similar SPI configuration?  I like that I can have any bit length of data on the SPI bus, this will be useful for JTAG data. Have you worked directly with JTAG data?

HSPI Frustrations

I cut the traces going to the SPI RAM that I figured out were wrong last week. On the PCB at U3 Pin3, and Pin5 I cut one trace each and on Pin6 I cut two traces.  Most of these cuts are under the chip so I made them before re-soldering the chip to the board.  Of course then I added wire jumpers to route the signals to match the schematic. I then added pins to connect the logic analyzer to.

Cut traces Chip Installed Rewired SPI





Last week I was looking at the the SPI driver code by MetalPhreak on Github as well as his blog to figure out how to send commands and data out and receive back over SPI.  The register SPI_USER is used to configure the SPI, SPI_USER1 sets up data length, SPI_USER2 sets up the command length and Command data.  Additionally, there is an SPI_Address Register that I can use to select an address in the RAM. Finally SPI_W0 to SPI_W15 are the data packet working registers — This is where I will put my “Hello World”.  I can write or read up to sixteen 4 byte words.  My RAM is organized in bytes so I will be putting bytes of data into the working registers.

So for initial setup I think the following should work for writing data.




The following should work for reading data:



The CS_SETUP and CS_HOLD parameters affect chip select timing. Setup is before clocking out data, hold is after.  Some chips need extra set up or hold time.  It will not affect performance significantly so I just turned each on by default so I don’t have to wonder if this RAM chip needs them. USR_COMMAND tells it to send the data in the command register. Likewise USR tells it to send the data in the Address Register.  USR_MOSI tells the hardware that there is data in the Working registers to be sent out. USR_MISO tells the hardware that I want to receive data into the working registers.

For the SPI RAM I want to send a command(Write), An Address(0), and 12 Bytes of data(“Hello World”).  This totals 16 bytes of data to be sent on the SPI bus:

1 byte for the command: 0x02
3 bytes for the address: 0x000000 and
12 bytes of data: “Hello World” including 1 trailing NULL.

For my first attempt to write to RAM I chose to hard code the instructions.
In my SPIRam.c File I had already created a function called writeRam.  I deleted it’s contents, and added the stuff to make this work.
void writeRam(char data[], int length){


((7&SPI_USR_MISO_BITLEN)<<SPI_USR_MISO_BITLEN_S)| //Ignored for write
((23&SPI_USR_ADDR_BITLEN)<<SPI_USR_ADDR_BITLEN_S)); //address is 24 bits A0-A8

WRITE_PERI_REG(SPI_ADDR(HSPI),(uint32) 0x000000<<(32-24)); //write 24-bit address

(((7&SPI_USR_COMMAND_BITLEN)<<SPI_USR_COMMAND_BITLEN_S) | 0x02)); // Write Command

WRITE_PERI_REG(SPI_W0(HSPI),0x48656C6C); //"Hell"
WRITE_PERI_REG(SPI_W1(HSPI),0x6F20576F); //"o Wo"

SET_PERI_REG_MASK(SPI_CMD(HSPI), SPI_USR);// Tell hardware to do it.

After fixing my typos, I compiled and loaded the code.  The system booted without hanging. then I hooked up the logic analyzer to look for the data being transferred and it didn’t work. The Chip select line(U3 Pin1 ) doesn’t go low for a frame of data.  This means the chip never gets selected to write to.

With several hours of tinkering, I finally have it close, but it is still not working correctly.  The CS0 pin is being activated when the HSPI CS pin is being activated.  So I am getting the correct sequence of bytes when HSPI CS is low, but CS0 is low at the same time.

HSPI almost working

I have not posted a new copy of the software, I don’t believe it is useful to anyone yet.  It isn’t even useful to learn from.

Have you had similar frustrations? Have you gotten the HSPI to work in overlap mode?

Please leave a comment below.